Jiahao Sun, CEO and Founder of FLock.io – Interview Series

Jiahao Sun, CEO and Founder of FLock.io, has led the company since April 2022. Prior to founding FLock.io, he spent many years in AI-driven roles — including as Director of Artificial Intelligence at a major financial firm — and served as a Research Fellow at a top university studying knowledge-graph generation, NLP, and machine-learning for financial prediction. His extensive background in both enterprise AI and academic research makes him well-positioned to lead a project aimed at reinventing how AI is built and deployed.

FLock.io is a decentralized AI platform that combines federated learning with blockchain. Rather than centralizing data, it enables communities to collaboratively build, train, and deploy AI models while keeping information private on local devices. The system offers incentives for contributors who provide compute, data, or validation, and distributes ownership and governance across the network to ensure transparency, security, and long-term resilience.

You founded FLock.io after years working in advanced AI research and financial prediction at Imperial College as well as leading AI innovation at RBC Wealth Management. What originally sparked the idea to build a decentralised AI platform, and what gap did you see that traditional AI architectures couldn’t address?

The idea for FLock came from seeing the same problem in both my academic research and finance work. Valuable data was locked away, and organisations could not use AI together without giving up privacy or control. Traditional AI systems are built around large central providers, and that limits who can participate and how widely the benefits can be shared. I realised that a decentralised approach could solve this. With new tools like blockchain, federated learning and privacy-preserving technology, people and companies can contribute data or computing power safely, while keeping ownership. FLock was created to make AI more open, more secure, and something that everyone can help build, not just a few major players.

Your background includes knowledge graph research and zero-shot learning for financial prediction. How has that experience shaped the way FLock.io approaches model training, validation, and data integrity on-chain?

I have learnt through experience that financial prediction is an inherently high-stakes domain where data integrity, model explainability, and the ability to generalise from limited data sets are paramount. In finance, the source and quality of data are everything. We translated this need for verifiable provenance into our on-chain architecture at FLock. We are structuring a decentralised network of data silos, allowing the AI to learn the collective knowledge without ever centralising the raw data itself. This allows us to build powerful models in sensitive sectors like healthcare and finance, where the data must remain distributed and private.

FLock.io’s AI Arena has already produced more than 9,000 fully trained machine learning models in under a year. What does this level of activity reveal about the demand for decentralised AI, and why do you think developers are choosing community-driven training over centralised alternatives?

This level of activity reveals two critical insights about the current state of the AI market. First, it confirms the massive demand for decentralised AI infrastructure. Developers are actively seeking alternatives to the centralized cloud model because it’s structurally brittle, expensive, and creates a single point of failure for data. Unlike centralized platforms where the participants own the model, in the AI Arena, the model developers and contributors maintain ownership and are rewarded with FLOCK tokens when their models are used.

Decentralised AI relies heavily on incentives to attract validators, delegators, and node operators. How does the FLOCK token design align incentives across the ecosystem, and what economic behaviors have you observed now that there are over 55,000 token holders?

The FLOCK token is our governance and utility engine which helps us ensure that the incentives of every participant are perfectly aligned with the network’s long-term health and the quality of the models produced. Our design is built on a clear principle to reward productive, verifiable contributions that drive collaboration among data owners, compute providers, validators, and delegators. For instance, compute providers and node operators earn FLOCK for contributing their computational resources to model training. Validators are rewarded for ensuring the security and accuracy of model updates, while delegators, who stake FLOCK to support them, earn a share of those rewards. We are seeing a sustained commitment from our community, with a high staking ratio indicating a preference for long-term network participation over short-term trading.

FLock.io has secured partnerships with the United Nations Development Programme, Alibaba, the NHS, and other major institutions. What convinced these organisations to adopt a decentralised AI framework instead of traditional cloud-based training?

These partnerships have helped us implement solutions to the challenges traditional, cloud-based training cannot address. Our framework, which combines federated learning with blockchain, allows AI models to be trained directly on local servers. Only the aggregated, anonymised model insights are shared to the blockchain. All transformative AI breakthroughs require pooling data insights across institutions that would never trust a central intermediary. Our work with the UNDP is a prime example, where with NGOs, we collaboratively build AI models to attain sustainable development goals. Our blockchain-governed framework provides the transparency and community consensus necessary for ethical AI development at a global scale.

Privacy-preserving AI is becoming essential in sectors like healthcare, public governance, and scientific research. How does FLock.io ensure that sensitive data can be used to train high-value models without ever being exposed or centralised?

FLock’s core technology brings together the best of federated learning and blockchain technology, which ensures that sensitive data can be used to train high-value models without ever being exposed or centralised. Instead of collecting vast amounts of sensitive data into a single, central location, federated learning lets multiple entities collaboratively train a model by bringing the training process directly to the data – then just collecting the updates, not the raw data. Add blockchain and it becomes yet more privacy-preserving and scalable through a system of crypto incentives.

For instance, we enable breakthroughs in areas like drug discovery and public health by allowing researchers to tap into a vast pool of real-world data, all while maintaining full data control and compliance.

As both crypto and AI evolve rapidly, how do you see the intersection of blockchain-based infrastructure and machine learning developing over the next five years, and what role will decentralised model marketplaces like FLock.io play?

The monopoly of centralised AI companies won’t last forever, as the world wakes up to the powerful potential of more collaborative, sovereign AI. Decentralised frameworks will become the default infrastructure for any sensitive AI application, for their local training and data control. As compute power becomes increasingly distributed and commoditised through decentralised networks, the true value will shift to the data and the models themselves. We are building the infrastructure for an AI future where privacy is a guarantee, not a trade-off, and blockchain will be the go-to technology to ensure it.

The AI Arena uses a task-based structure where communities collaboratively train and validate models. How does this competitive-collaborative environment improve model performance, and what new use cases are emerging from it?

The competitive element incentivises our network of machine learning engineers to submit the best-performing model tasks, earning rewards based on the accuracy and efficiency of their submissions. Meanwhile, the collaborative consensus mechanism improves transparency by filtering biases from centralised entities. It’s designed to produce more robust and specialised AI models more efficiently than traditional centralised models.

Through this approach, FLock enables diverse communities to develop purpose-built AI models, offering bespoke solutions tailored to specific needs. We’ve found this structure truly excels in diverse, global and niche domains where preserving privacy is paramount. This includes healthcare, financial services and supply chain logistics, where sensitive data can be analysed locally, without ever pooling the raw information into a central database, ensuring maximum security.

You previously built NORA at RBC—one of the leading AI systems for wealth management. How does developing AI in a regulated financial environment compare to building decentralised AI infrastructure where transparency and openness are built into the core?

Interestingly enough, the technical fundamentals aren’t so different. Every model needs to be validated and audited for inherent biases and requires a tremendous amount of effort from skilled engineers. The difference, however, is in the cultural mindset fuelling the work itself.

In wealth management, the focus is on a single, highly controlled model that guarantees regulatory compliance but lacks agility. Risk mitigation and compliance are at the forefront of every activity, whereby decentralised AI, naturally, flips that script. Instead, transparency and openness are considered as fundamental design principles as opposed to afterthoughts. This openness attracts a more diverse community of contributors, which inherently makes the final models more resilient to the biases and single points of failure that can plague models built by a single, centralized team.

FLock.io was the only decentralised AI training project named to the CB Insights AI 100. What do you believe differentiates decentralised AI from the broader wave of AI acceleration happening today?

What differentiates decentralised AI from the current wave of innovation focused on computational scale is the shift in focus to governance, privacy, and economic alignment. The core principles of decentralisation, transparency and openness directly challenge the stature of the few centralised players that currently dominate the AI space to ensure fairer economic opportunities, greater global collaboration and true data sovereignty and privacy.

Being the only decentralised AI training project named to the CB Insights AI 100 completely validates our value proposition and core thesis. It shows that decentralisation is the next and necessary step to facilitate innovation that will enable AI to be used for good.

Thank you for the great interview, readers who wish to learn more should visit FLock.io.